On Mon, Feb 18, 2008 at 04:22:42PM -0500, Chris Rapier wrote: > As Ben said - we aren't wedded to the idea of threads. They were a > useful path to take in order to prove the usefulness of some sort of > parallelization in OpenSSH. I think we've proved its usefulness (and > believe it or not, a lot of people are using SSH for bulk file transfer) > and as such we're more than happy to think about other methods. In fact, > I think Ben is working on an IPC implementation now. Honestly, I've > never cared about the code getting into the OpenSSH base. I've only > really been concerned about the underlying concepts. > > So if anyone does have any thoughts you might happen to have on > implementation issues they would be appreciated very much. What would be > really interesting is to have a bake off but I expect most people, even > though they might feel strongly about this, don't have the time to > invest in any coding.
I wonder where the perceived bottleneck is. I mean, you have two boxes connected by ethernet (whatever speed), and you're running a sftp bulk file transfer. What is the limiting factor? Are the boxes less than 20% idle? Is the nework saturated or is there room for more throughput? In other words, with a multiple-CPU box, how much would threads help? Within the whole ssh section, where does the CPU spend its time? Is it crypto or is it in shuffling network packets? Would offloading the crypto to a separate process (and therefore processor) help? Doug.
